Rethinking Loss Functions for Fact Verification

This is the official implementation of the following paper:

Yuta Mukobara, Yutaro Shigeto, Masashi Shimbo. Rethinking Loss Functions for Fact Verification. EACL 2024

[ arXiv | ACL Anthology ]

The following shows how to clone a repository including even submodules.

git clone --recursive [email protected]:yuta-mukobara/RLF-KGAT.git

A Docker environment for thunlp/KernelGAT

Build a docker image

make docker-build

Download data and checkpoint

BERT based models and checkpoints used for training and RoBERTa based models and checkpoints can be downloaded with the following command.

make download

If the above command did not download successfully, you can download from Ali Drive in thunlp/KernelGAT.

All data and BERT based chechpoints: Ali Drive

RoBERTa based models and chechpoints: Ali Drive

Preprocess

Set up for train, test and eval.

make prepro

Usage

Train

make kgat

Hyperparameter

• comp (None, all, srn, sr) OVR refers to XE in the paper.
• nl_coef (float) nl_coef corresponds to $\lambda$ in the paper.
• imb (store_true) with/without imbalanced learning
• beta (float) beta refers to the Weighting $\beta$ in the paper.

Test

make test

Evaluate

make eval

Citation

If you use this code, please cite our paper:

@inproceedings{mukobara-etal-2024-rethinking,
    title     = {Rethinking Loss Functions for Fact Verification},
    author    = {Mukobara, Yuta and Shigeto, Yutaro and Shimbo, Masashi},
    booktitle = {Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)},
    month     = {03},
    year      = {2024},
    publisher = {Association for Computational Linguistics},
    url       = {https://aclanthology.org/2024.eacl-short.38},
    pages     = {432--442}
}

Contact

If you have questions, suggestions and bug reports, please email:

[email protected]